0/1

2A1 Time Frequency Analysis: Fourier Series and Transforms

8 Lectures Hilary 2011 Rev February 2011 Course Page

David Murray [email protected] www.robots.ox.ac.uk/$dwm/Courses/2TF

Grey Book
Complex Fourier series: evaluation of complex coecients for periodic functions; inversion relationship; the idea of spectra. Fourier transform: derivation of transform from Fourier series; inverse transform; convolution integral; impulse response functions; proof and use of duality; convolution and Parsevals theorems. Introduction to sampling and reconstruction, including the sampling theorem and aliasing. Introduction to random processes.

Lecture Content
1. 2. 3. 4. 5. 6.

From Signals to Complex Fourier Series From Complex Fourier Series to the Fourier Transform Convolution. The impulse response and transfer functions Sampling, Aliasing Power & Energy Spectra, Autocorrelation, and Spectral Densities Random Processes and Signals

The content will ow from one lecture slot to another. If there is any time left over, we will go through examples.

0/2

References
These notes are not meant to be comprehensive. Fourier Analysis is a topic where a good book with decent diagrams and examples can make a dierence. So | do read promiscuously. This list will be updated on the website with comments.

 Kreyszig,  Riley, Hobson and Bence,

neering

Advanced Engineering Mathematics 6th ed

, Wiley

, CUP.

Mathematical Methods for the Physics and Engi-

 Baher  Denbigh,

Analog and Digital Signal Processing

Wiley, 1991.

System Analysis and Signal Processing: with emphasis on the use

of MATLAB

, Addison-Wesley, 1998.

 James,  Oppenheim, Willsky and Young,  Newland

Longman, 1993.

Advanced Modern Engineering Mathematics

Addison-Wesley, 1993. , Prentice-Hall, 1984. ,

Signals and Systems

Introduction to Random Vibrations, Spectral and Wavelet Analysis

 Papoulis

Hill, 1991.

Probability, Random Variables and Stochastic Processes

McGraw-

Course WWW Pages

Pdf copies of these notes (including larger print versions), pdf copies of the lecture slides, the tutorial sheets, FAQs etc will be accessible from

www.robots.ox.ac.uk/$dwm/Courses/2TF

Just the notes and the tute sheets get put on weblearn.

Topic 1 From Signals to Complex Fourier Series

1.1 Linear systems and frequency

By this stage, you will have realized how important linear systems and their analysis are to engineers. In a linear system | one described by a linear dierential equation of some order | the response y (t ) occurs at the same frequency as the input x (t ), and if the input's amplitude is changed by some factor, the output's amplitude will change by that same factor. When faced with a non-linear system, engineers often linearize the system by considering incremental inputs and outputs that occur around a xed operating point. The use of small signal equivalents in transistor circuits is an obvious example.

1.1.1 You know about Frequency Response

You will have an understanding too of the importance of the frequency response of a system, be it mechanical, electrical, or whatever. That response is described by the transfer function H (!), which may be complex, representing both the change in phase and amplitude between input and output. Strictly speaking, the transfer function relates inputs and outputs in the frequency domain

Y (! ) = H (! )X (! ) ;
but so far you have thought of X (!) and Y (!) as phasor representations of harmonic input and output at a single frequency. At a single frequency, the temporal input and output are related to the frequency representations by

x (t ) = Re X (!)ei!t ;

y (t ) = Re Y (!)ei!t = Re H(!)X (!)ei!t

As a concrete example, consider as input the voltage x (t ) = V0 cos !t applied to an inductor L in series with a resistor R, and as output the voltage y (t ) across R.
1

1/2

The input phasor, transfer function and output phasor are, respectively

V0R R ; X (!) = V0 ; H(!) = R + j!L ; Y (!) = R + j!L

Notice that although we can write Y (!) = H (!)X (!) there seems to be no equivalently crisp operation involving y (t ) and x (t ). Linear systems obey the principle of linear superposition. It says that if the input is a linear combination of signals, the output is the same linear combination of the individual outputs x (t ) = 1x1(t ) + 2x2(t ) + : : : A y (t ) = 1y1(t ) + 2y2(t ) + : : : : If the x1(t ) etc are harmonic signals at frequencies !1 etc, the output must be y (t ) = 1Re H(!1)X (!1)ei!1t + 2Re H(!2)X (!2)ei!2t + : : : : A rather more elegant way of thinking about this is to write the discrete frequency spectrum of x (t ) as X (!) = 1X (!1) + 2X (!2) + then Y (!) = 1Y (!1)+ 2Y (!2)+ = 1H(!1)X (!1)+ 2H(!2)X (!2)+ :

1.1.2 You know about Superposition

1.1.3 You know about Fourier Series

This is remarkable, but would be utterly arcane were it not for an amazing property of (most) periodic signals, viz: A periodic signal of angular frequency !0 can1 be represented as the sum of a set of harmonic signals at frequencies !0, 2!0, 3!0, and so on. These sums of harmonic waves are Fourier Series. For example, the Fourier series of a unit square wave with a zero at t = 0 and period T = 2=!0 is 1 1 4 1 4 sin n!0t : f (t ) =  sin !0t + 3 sin 3!0t + 5 sin 5!0t + =  n n odd Now, armed with a system's transfer function H (!), the principle of linear superposition, and this and similar Fourier Series, you can work out the output of the system corresponding to the square wave, or any other periodic input. This is illustrated in Figure 1.1.
1 Provided

the \Dirichlet conditions" are satised. See later.

1/3

X ( ) Fourier Series X (2 ) X (3 )

H ( ) H (2 ) H (3 )

Y ( ) Y (2 ) Y (3 )

Figure 1.1: The Fourier Series, the principle of linear superposition, and the transfer function, allow one to compute the output for any periodic input.
1.2 The gap in our knowledge

Unfortunately, not all inputs are periodic. Can Fourier analysis help us? Well, if the signal x (t ) is of nite duration we might make it periodic by \pretending" it repeats. However, this does not help with general signals of innite duration. Another way of perceiving the gap in out knowledge is to realize that well-behaved periodic functions give rise only to discrete frequency spectra. For example, the unit square wave (Figure 1.2(a)) has the frequency spectrum shown in Figure 1.2(b), where the components are in the ratio 1 : (1=3) : (1=5) : (1=7) : : : :. However, instinct tells us that there must be signals that have a continuous frequency spectrum as sketched in Figure 1.2(c). Nothing we know yet about Fourier analysis would allow the analysis of a continuous spectrum.
f(t) F( ) F( ) +1/3 +1/5 T= 2 / 0 (a) 0 3 0 5 0

+1

(b)

(c)

Figure 1.2: A periodic function (a) gives a discrete frequency spectrum (b). (c) A continuous frequency spectrum cannot be derived from a periodic function.

1/4
1.3 Fourier Transforms

We shall see that Fourier transforms provide a method of transforming innite duration signals, both non-periodic and periodic, from the time domain into the continuous frequency domain. In fact, they provide an entire language with which to work and think in the frequency domain. The language involves a deal of new vocabulary and several new mathematical techniques, like convolution, correlation, modulation, sampling, spectral density,  -functions and so on. Most of these techniques depend at the lowest level on integration. The integrals can look daunting, but it is important to rise up onto the next level | so that you can say \a signal's power spectral density is the Fourier transform of its autocorrelation". Success will come if you practice the mathematics, but also practice xing the concepts in your head by using simple physical examples.
1.4 Joseph Fourier

If at any point you become really cross, just remember that Fourier seems a really jolly chap. He had an interesting life not only in mathematics but in politics in France during Napoleon's time. Read his biography (and those of many other mathematicians) at the website maintained by the Department of Maths and Computer Science at St Andrew's University. Follow the links from wwwhistory.mcs.st-and.ac.uk/$history/
1.5 Signals

You will have noticed that the emphasis of our introductory discussion drifted from systems, and towards signals. Indeed, this course could be titled \An introduction to analogue signal processing". Let us start by dening various signal types. A signal might be a function of one or of several variables. Space and time are very common variables, but here we will tend to stick to one variable, and choose time t more often than not. Again more often than not, we will think about electrical signals | but do remember that the variation in temperature during the day is just as much a signal as the voltage output of a thermocouple sensing it. An Analogue signal is one whose amplitude covers a continuous range. It may be bounded (e.g. 0 5 V) but it can just as easily take the value 2:0572941675975 V as 2:0572941675974 V, and indeed anything between! A continuous-time analogue signal is one that has a value for a continuous sweep of values of its parameter, t . Some examples are shown in Figure 1.4. Notice the

1/5

Figure 1.3: Jean-Baptiste Joseph Fourier 1768-1830.

value can be zero, but it is dened as zero; and notice too that a continuous-time signal does not have to be a continous function.
f(t) f(t) f(t)

Figure 1.4: Examples of continuous-time analogue signals.

times only. Typically these will be at regular intervals, arising from regular sampling. Examples are shown in Figure 1.5. Note that rather than f (t ), this type of signal is labelled f (nT ), where n is an integer, and T is the sampling interval. A digital signal is one that is by denition sampled, but whose amplitude can only take one of a discrete set of values represented by some binary coding scheme. Suppose we used 4 bits. This could represent f0; 1; : : : ; 15g V, or f0:2; 0:1; 0:0; : : : ; 1:3g V, or f0; 1; 4; 9; : : : ; 225g V; but, using the rst example, we cannot properly represent any value between 0 and 1 V. In this course we are not concerned with digital signals, and consider only analogue, continuous-time and discrete-time.

A discrete-time analogue signal is one that has an analogue value at certain

1/6

f(t)

Sampling Period T

f(nT)

Sampling Period T

Figure 1.5: A continuous time signal f (t ) sampled at intervals of T to general a discrete-time signal f (nT ).

A causal signal is one that is nite only for t > 0 | ie, f (t ) = 0 for all t < 0. A casual signal is one that has been mis-spelled. See above. A deterministic signal is one that can be described by a function, mapping, or
some other recipe or algorithm. If you know t , you can work out f (t ). We shall be interested in deterministic signals for much of the course. A random signal is determined by some underyling random process. Although its statistical properties might be known (e.g. you might know its mean and variance) you cannot evaluate its value at time t . You might by able to say something about the likelihood of its taking some value at time t , but not more. We will come to think about random processes towards the end of the course.
1.6 Orthogonal basis functions - Revision

Before revising the Fourier series, we think about orthogonality and functions, taking a scenic meander via vectors. Consider three vectors v1, v2, v3, which are of dierent lengths but lie at right angles to each other. They form a set of orthogonal basis vectors in 3D. It should be evident that any 3D vector f can be described by a unique linear combination of the vectors How would you nd these unique coecients for a particular f ? To nd A1, take the scalar or inner product of both sides with v1 and then exploit orthogonality which tells you that v2 v1 = 0 and v3 v1 = 0, so

f = A1v1 + A2v2 + A3v3 :

f v1 = A1v1 v1 + A2v2 v1 + A3v3 v1 ;

1/7

that

(The denominator is important | v1 and so on were not unit vectors.) In the early 19th century it was realized that it was possible to treat functions f in the same way and to make them up from a set of orthogonal basis functions. Pretend there is a set of orthogonal \v -functions", so that, similar to the vector case,

A1 = v v1 : 1 1

f v

f (t ) = A1v1(t ) + A2v2(t ) + A3v3(t ) + :

How do we nd A1 and so on? First we need to dene the equivalent of the scalar or inner product between functions | but for now just assume we can, and we denote it hf ; g i. Then, by analogy

hf ; v1i = A1 hv1; v1i + A2 hv2; v1i + : But hv2; v1i = 0 and so on, so that hf ; v i hf ; v i and An = hv ; vn i : A1 = hv ; v1 i 1 1 n n
There are a considerable number of famous sets of orthogonal basis functions, many derived by French mathematicians, but it was Fourier who noticed that the cosines and sines made up such a set which were then capable of representing periodic functions.
1.7 Fourier Series

Fourier found that a periodic function f (t ), with period T , can be written as a sum of cosine and sine functions of the fundamental frequency ! = 2=T and its harmonics 2!, 3!, etc
I I 1 f (t ) = 2 A0 + An cos(n!t ) + Bn sin(n!t ) : n=1 n=1

The expressions for An and Bn are found from the orthogonality conditions as

1/8

An = T Bn = T

2 2

+T=2

T=2 T=2

+T=2

f (t ) cos(n!t )dt n = 0; 1; : : : A A0 = T f (t ) sin(n!t )dt n = 1; 2; : : :

+T=2

T=2

f (t )dt

To derive the stated expressions for the coecients, we must rst dene the inner product for these functions. It is dened by the integral over a period, divided by the period. Then, we should demonstrate that the basis functions, cn = cos n!t and sn = sin n!t , are orthogonal. In other words, we must show that the inner products hcn ; cm i and hsn ; sm i are zero for m T= n, that hcn ; sm i = 0, and that only hcn ; cn i and hsn ; sn i are nite. One nds 0 1=2 T T=2 cos m!t cos n!t dt = 1 0 1 T=2 1=2 T T=2 sin m!t sin n!t dt = 0 1 T=2 cos m!t sin n!t dt = 0 1
T=2

1.7.1 Finding the coecients

m T= n m=n>0 m=n=0 m T= n m=n>0 m=n=0

T=2

The only complication is that we have both cn and sn as in the basis set, so we need two coecients with the subscript n. Calling these An and Bn we nd
1 T=2 h f ; cn i 2 T T=2 f (t ) cos n!t dt An = hc ; c i = 1 T=2 = T n n T T=2 cos n!t cos n!t dt

T=2

T=2
T=2

f (t ) cos n!t dt

and
1 T=2 h f ; sn i 2 T T=2 f (t ) sin n!t dt Bn = hs ; s i = 1 T=2 = T n n T T=2 sin n!t sin n!t dt

T=2

f (t ) sin n!t dt

1/9

1.7.2 A more explicit approach

1
T=2

If you felt there was sleight of hand in the above, you may prefer to write down the series, and then, to nd the Bm for example, multiply it by sin(m!t ) and average over a period:
T T=2 f (t ) sin(m!t )dt =
I I 1 A0 + An cos(n!t ) + Bn sin(n!t ) sin(m!t )dt T T=2 2 n=1 n=1
1
T=2

The three terms on the right are 1 T T=2 2 A0 sin(m!t )dt = 0 1

T=2

T=2

T=2 n=1 T=2 I T=2 n=1

An cos(n!t ) sin(m!t )dt = Bn sin(n!t ) sin(m!t )dt =

I I

n=1 n=1

An T

1 1

T=2

T=2
T=2

cos(n!t ) sin(m!t )dt = 0 sin(n!t ) sin(m!t )dt = Bm 1 2

Bn T

T=2

Hence, in agreement with the earlier statement,

T=2

T=2

f (t ) sin(m!t )dt = Bm 2

Similarly, to obtain Am we would multiply by cos(m!t ) and average over one period.

1/10
1.8
|

|
2

Examples

Eg 1: Square Wave
+T=2

The period is T , so ! = 2=T . 2

0

Am = T

T=2

f (t ) cos(m!t )dt = T

T=2

1 cos(m!t )dt + 1 sin(m!t )dt +

T=2
0

1 cos(m!t )dt = 0

Bm = T
= =

2 4 2
0

+T=2

T=2
T=2

f (t ) sin(m!t )dt = T
4 1 2

T=2
0

T=2

1 sin(m!t )dt

m

[ cos(m) + 1] =

2 sin(m!t )dt = T m! [ cos(m!t )jT= 0

m

(1)m+1 + 1

This gives the series stated earlier:

f (t ) =  sin(!t ) + 3 sin(3!t ) + 5 sin(5!t ) + : : :

Eg 2: Triangular waveform

Period is 2 hence ! = 1.
0

2 + 1 Am = 2 f (t ) cos(m!t )dt =   1 2

t cos(mt )dt +


0

t cos(mt )dt

Use integration by parts with u = t , dv=dt = cos mt :

Am = m2 [2 + 2 cos m] = m2 [1 + (

2 Bm = 2
0

1)m ]

for m = 0 for m EVEN = 0 2 4=(m ) for m ODD

 4 1 1 f (t ) = 2 +  cos(t ) + 9 cos(3t ) + 25 cos(5t ) + : : :

t sin(mt )dt +


0

t sin(mt )dt = grind = 0 :

1/11
1.5 1

f(t)

1
y(t)

0.5 0 0.5 1 1.5 1 1.5 1 0.5 y(t) 0 0.5 1 1.5 1

T/2 1 Period T
1.5 1 0.5 y(t) 0 0.5 1 1.5 1

+T/2

0.5

0 t

0.5

0.5

0 t

0.5

0.5

0 t

0.5

Figure 1.6: The FS of a square wave built up over 1,3,5 terms; then 11 and 101 terms; then 1001 terms. The slight \chip" at the discontinuity is a result of the Gibbs' phenomenon, discussed later.

The period is T , so ! = 2=T . The function is even, so only An coecients exist.

|

Eg 3: Train of tophats

A0 = T

f (t ) dt = T T=2

T=2

2a

An = T

4
0

a=2

cos n!t dt = =

T n!
2

4 1

2 sin n!t ja= 0

n sin(n!a=2)

a I 2 So f (t ) = + T n=1 n sin(n!a=2) cos n!t Of interest later are the values of the An . Suppose we set the (on/o) ratio to be  = a=T , sin(n) An = 2 (n)
If the on/o ratio  = 1= then the An are taken from the sin(x )=(x ) curve as shown in Figure 1.8(b). If we reduce , here by 1/2, then the An values are sampled from the curve more closely, as in Figure 1.8(c).

1/12

f(t)

Period 2
3 2.5 2 1.5 1 0.5 0 -4 -3 -2 -1 0 1

t
"tri.5" "tri.11"

Using 1 then 5 then 11 terms Figure 1.7:

1.9 Not any periodic function ...

There is a set of conditions, known as the Dirichlet conditions, that determine whether or not a function can be expressed as a Fourier Series. A function must 1. be periodic, or be of nite extent so that it can be made periodic by extension; 2. have only a nite number of discontinuities within a period; 3. the discontinuities must be of nite size. 4. nite number of maxima and minima So, we can nd the FS of the function f (t ) = et , 1 t < 1, in Figure 1.9(a), but not of f (t ) = 1=t for 1 t < 1 in Figure 1.9(b).

1/13

f(t)
1

a/2 a/2 Period T

1.0

t
1.0

sin(x)/(x)

sin(x)/(x)

x 0 5 10 0 5

x 10

Figure 1.8:
1.10 Fourier Series at Discontinuities

Provided the Dirichlet conditions are satised, at a point of discontinuity in the orginal function, a Fourier series converges to 1 FS(t ) 3 (f (t) + f (t+)) 2 where f (t) is the value of the signal f (t ) just below the discontinuity, and f (t+) that just above. We have already seen this in the square wave example. We have also seen is that we require a large number of terms in the series to faithfully reproduce the function at a discontinuity.
1.11 Symmetry properties

The task of deriving series coecients is made a little easier by exploiting symmetries in the signal f (t ) and the basis functions sin(n!t ) and cos(n!t ). The sine basis functions all have have odd 1=2-wave symmetry; ie, sin(n!t ) = sin(n!t ) (Figure 1.10). If the signal f (t ) is even, then all integrals
T=2

T=2

f (t ) sin(n!t )dt

1/14

1 1 +1

+1

(a)

(b)

Figure 1.9: The function 1=t in (b) has an innite discontinuity at t = 0.

must be zero. Therefore

 an even signal is contains only cosine terms; and, similarly,  an odd signal contains only sine terms.
One also notes that for an even, then odd, signals f (t )
T=2

T=2
T=2

f (t ) cos(n!t )dt = 2 f (t ) sin(n!t )dt = 2

T=2
0

f (t ) cos(n!t )dt

T=2
0

T=2

f (t ) sin(n!t )dt
sin 2 t

sin t

1/4 wave 1/2 wave 1/4 wave

reflection reflection reflection

1/4 wave 1/2 wave 1/4 wave

reflection reflection reflection

Figure 1.10: Basis function sin(!t ) has odd 1/2-wave, even 1/4-wave symmetry. sin(2!t ) has odd 1/2-wave, odd 1/4-wave symmetry.

Further use can be made of symmetries about the 1/4-wave points. Any sin(n!t ) with n-even has odd symmetry about these points. Thus if a signal f (t ) has even symmetry about the 1/4-wave points, any n-even sine terms will vanish. However

1/15

if the signal's symmetry is odd about the 1/4-wave points, the n-odd sine terms vanish. Similar arguments can be made for the cosine series. The square wave and triangular waves (Figure 1.11) are examples, but do note that the signal f (t ) does not have to have 1/2-wave symmetry to exploit 1/4-wave symmetry. One could consider higher symmetries, but they get increasingly dicult to recognize.

f(t)

f(t)

T/2 1 Period T


+T/2

t Period 2


4 sin(!t ) + 1 sin(3!t ) + 1 sin(5!t ) + 3 5

4 1 1 2 +  cos(t ) + 9 cos(3t ) + 25 cos(5t ) + : : :

Figure 1.11:
1.12 Completing Functions

Suppose you need derive the Fourier series of a non- time-continuous, non-period function. For example, f (t ) = t for 0 t 1. First, you MUST make the function periodic | but exactly how is a matter of choice. Figure 1.12 shows three possible ways of completing the example function. Which is best? Because the cosine series has no discontinuities it will require fewer terms to make an overall decent approximation. However, the kink at t = 0 will not be accurate. If a good approximation with few terms is required close to t = 0, it is probably best to use the sine series completion.
1.13

Something rather curious occurs in the FS at a discontinuity. Residual oscillations lead to an overshoot, whose size is a characterstic of the underlying function. (It can be large! For a square wave of amplitude A, the overshoot is around  = 0:18A.) As more and more terms are added to the series, the oscillations get squeezed into a shorter and shorter region around the discontinuity, but their characteristic amplitude remains constant! This eect is known as the Gibbs phenomenon.

The Gibbs phenomenon

1/16
Defined

Completed as sine

as cosine

as mixed sin, cos

Double period sine completion

Figure 1.12:
1.5
1 "gibbs.1001" "gibbs.10001"

1
0.5

0.5 y(t) 0 0.5 1 1.5 1

0

-0.5

-1

0.5

0 t

0.5

-1.5 0.98

0.985

0.99

0.995

1.005

1.01

1.015

(a)

(b)

Figure 1.13: The Gibbs phenomenon. (b) shows the discontinuity of (a) at a ner time scale, but with more terms added.
1.14 Mean square values of Fourier Series Parseval's Theorem

The instantaneous power in a signal is proportional to the its modulus squared. For a periodic signal we can derive the average signal power by integrating over a period.

Ave signal pwr

= =

+T=2

T=2

jf (t )j2dt

+T=2

T=2

1 A + An cos(n!t ) + Bn sin(n!t ) dt 2 0 n=1 n=1

1/17

At rst this looks nightmarish, because the squaring introduces an innite number of cross terms, on the bottom line of the next expression.
2 1 2 2 2 2 2 2 2 A0 + A2 1 cos !t + A2 cos 2!t + : : : + B1 sin !t + B2 sin 2!t + : : : + T T=2 2 A0 A1 cos !t + : : : + A0 B1 sin !t + : : : + 2A1 B1 cos !t sin !t + 2A1 B2 cos !t sin 2!t + : : :g dt
1

+T=2

However, when you integrate over cross terms, orthogonality will get rid of all the terms on the bottom line! So, we are left with the simple result that the mean square is

Ave pwr = T

+T=2

T=2

jf (t )j dt = 1 A 2 0
2

1 I 1 I 2 + (An ) + (Bn )2 2 n=1 2 n=1

A good way to remember this is as 1 Mean Square = (d:c: amplitude)2 + 2 (a:c: amplitude)2

which is true whether the signal is pure d.c. or single frequency a.c. The Root mean square value of a periodic signal is therefore

Root Mean Square =

1.15

1 A 2 0

1 I 1 I 2 (An ) + (Bn )2 + 2 n=1 2 n=1

circuit in (b). What is the average power dissipated?

12V

[Q] Suppose the full wave rectied voltage in Figure 1.14(a) is applied to the
v(t)
0 10 20 t (ms)
Figure 1.14:

Fourier and Parseval in an electrical circuit

5k

1 F

1/18

[A] From the diagram T = 102 s and !0 = 2=T = 200. By grinding, or from
HLT, we nd the signal v (t ), with t in seconds, is 24 2 2

v (t ) =  1 + 3 cos !0t 15 cos 2!0t + : : :

We know that power is dissipated in the resistor alone, so 1 24 Pave = R v = 5 103 
2

1 1 + 2
2

2 3

2 + 15

+ :::

The current drawn from the source at a single frequency ! is I (!) = Y (!)V (!), where Y is the admittance. But we have voltage components at ! = 0, ! = !0, ! = 2!0 and so on. We must work out the admittances at all these frequencies.

[Q] What is current drawn from the source?

1

Y (!) = R + j!C = (2 104) + j!(1 106) 4 10 (2 + j 0); ! 104(2 + j 2); = 104 2 + j = 104(2 + j 4); 100 4 10 (2 + j 2n);

dc ! = !0 = 200 ! = 2!0 = 400 ! = n!0

To avoid mixing functions of time with phasors, it makes sense to rewrite v (t ) as the real part of

v (t ) = Re  1 + 3 ej! t 15 ej 2! t + : : :
0 0

24

Hence

i (t ) = Re

2 2 2 + (2 + j 2) ej 200t (2 + j 4) ej 400t + :::  3 15 48 104 2 = 1 + (1 + 2)1=2 cos(200t + 1)  3 2 (1 + 42)1=2 15 cos(400t + 2) + :::

24 104

with 1 = tan1  and 2 = tan1 2.

1/19
1.16 The Complex Fourier Series

You will have noticed while working out i (t ) that there was a rather unsatisfactory moment when we had to rewrite the Fourier series using an exponential representation. It raises the following question.

Would it be possible use an exponential, or complex, form of the Fourier Series from the outset?
It turns out | perhaps not surprisingly | that the set ein!t provides a set of orthogonal basis functions, and that a periodic function that satises the Dirichlet conditions and has period T = 2=! can be written as

The complex Fourier series (NB: n ranges from I to +I.)

f (t ) =
I
n=I

Cn ein!t

The inner product is dened as integration over a period with the complex conjugate, divided by the period, and the orthogonality conditions are simpler than before, 1
T=2

T=2

ein!t eim!t dt =

0 1

m T= n m=n

We determine the form of Cn using the orthogonality relationship in either the short hand or long hand ways. The short hand way says

Cm = heim!t eim!t i =

f (t ) eim!t

1 T=2 im!t im!t e dt T T=2 e

T=2 f (t )e

T=2

im!t dt

T=2

T=2

f (t )eim!t dt

1/20

The long hand version is not much longer. If f (t ) =

I
n=I 1 T=2

Cn ein!t
I

T=2

T=2

f (t )eim!t dt =
= =

T=2 n=I

Cn ein!t eim!t dt
ein!t eim!t dt

n=I Cm

Cn T

T=2

T=2

The coecients in the complex Fourier series are

Cm = T
1
T=2

T=2

f (t )eim!t dt

There is a third way of deriving the complex Fourier series. It is a bit feeble because it simply uses the relationship eim!t = cos m!t i sin m!t : The relationship between the C coecients and those for the non-complex Fourier series is (Am iBm )=2 for m > 0 for m = 0 Cm = A0=2 (Ajmj + iBjmj)=2 for m < 0
, where denotes complex conjugate. Also note that Cm = C m
1.17 Parseval and the Complex Fourier Series

This is asked as a question on the rst tutorial sheet, and I want you to think about it, rather than copy it down. Notes on this will appear later in the course to check your answers!

1/21
1.18 Summary

We have dened certain terms used to describe signals. We have reviewed how periodic signals can be represented as Fourier Series | linear sums of pure harmonic signals | and revised their properties. We have introduced the complex Fourier Series, which is often a more convenient representation to use when having to deal with phase shifts. Interesting and useful as the Complex Fourier Series is, there is nothing in it that addresses the gap in our knowledge, which is how to cope with non-periodic signals of innite duration; that is, those which have continuous spectra in the frequency domain. This is where we move next.